• deep learning
• detr
• document layout analysis
• faster r-cnn
• object detection
• transfer learning
• yolo

The research on object detection has come a long way and has in recent years become so efficient and reliable that we see many practical applications in simple detection of objects in pictures to advance segmentation in video recordings. State-of-the-art deep learning techniques that are designed to work on images, are adapted to the domain of digital documents. In this thesis work, an automatic method to extract meaningful information from documents is proposed. The study is based on the most recent object detection algorithms trained to detect various fields present in an Italian vehicle certificates, a document type that has not been considered in the literature. A custom dataset of 453 certificates images has been annotated. Three different models (YOLO, DETR, Faster RCNN) have been compared and evaluated on the mean Average Precision (mAP), common objects in context (COCO) evaluation metrics. Several experiments have been carried out, exploiting different combinations of data augmentation techniques. The obtained results are encouraging, leading to the conclusion that object detection is a viable method for information extraction.